When a reference appears in a function argument, the reference is
assumed to be valid for the entire duration of that function call; this
is called a stack protector [1]. Because of that, custom pointer types
whose destructor may invalidate the pointee (i.e. they are more similar
to Box<T> than &T) cannot internally use a reference, and must instead
use a raw pointer.

This issue is something that is often missed during unsafe review. For
examples, see [2] and [3]. To ensure that people don't try to simplify
RevocableGuard by changing the raw pointer to a reference, add a comment
to that effect.

Link: https://perso.crans.org/vanille/treebor/protectors.html [1]
Link: https://users.rust-lang.org/t/unsafe-code-review-semi-owning-weak-rwlock-t-guard/95706 [2]
Link: https://lore.kernel.org/all/aEqdur4JTFa1V20U@google.com/ [3]
Signed-off-by: Alice Ryhl <aliceryhl@google.com>
---
 rust/kernel/revocable.rs | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/rust/kernel/revocable.rs b/rust/kernel/revocable.rs
index db4aa46bb1216d18868dcbdcb0b7033a757306b1..5dd7c5a24809ac372cf600b1cf92c455901e42d6 100644
--- a/rust/kernel/revocable.rs
+++ b/rust/kernel/revocable.rs
@@ -221,6 +221,10 @@ fn drop(self: Pin<&mut Self>) {
 ///
 /// The RCU read-side lock is held while the guard is alive.
 pub struct RevocableGuard<'a, T> {
+    // This can't use the `&'a T` type because references that appear in function arguments must
+    // not become dangling during the execution of the function, which can happen if the
+    // `RevocableGuard` is passed as a function argument and then dropped during execution of the
+    // function.
     data_ref: *const T,
     _rcu_guard: rcu::Guard,
     _p: PhantomData<&'a ()>,

---
base-commit: 19272b37aa4f83ca52bdf9c16d5d81bdd1354494
change-id: 20250612-revocable-ptr-comment-7c82b47a503f

Best regards,
-- 
Alice Ryhl <aliceryhl@google.com>

On Thu, Jun 12, 2025 at 1:17 PM Alice Ryhl <aliceryhl@google.com> wrote:
>
> When a reference appears in a function argument, the reference is
> assumed to be valid for the entire duration of that function call; this
> is called a stack protector [1]. Because of that, custom pointer types
> whose destructor may invalidate the pointee (i.e. they are more similar
> to Box<T> than &T) cannot internally use a reference, and must instead
> use a raw pointer.
>
> This issue is something that is often missed during unsafe review. For
> examples, see [2] and [3]. To ensure that people don't try to simplify
> RevocableGuard by changing the raw pointer to a reference, add a comment
> to that effect.
>
> Link: https://perso.crans.org/vanille/treebor/protectors.html [1]
> Link: https://users.rust-lang.org/t/unsafe-code-review-semi-owning-weak-rwlock-t-guard/95706 [2]
> Link: https://lore.kernel.org/all/aEqdur4JTFa1V20U@google.com/ [3]
> Signed-off-by: Alice Ryhl <aliceryhl@google.com>

Applied to `rust-next` -- thanks everyone!

    [ Adjusted title. - Miguel ]

Cheers,
Miguel

On Thu Jun 12, 2025 at 1:17 PM CEST, Alice Ryhl wrote:
> When a reference appears in a function argument, the reference is
> assumed to be valid for the entire duration of that function call; this
> is called a stack protector [1]. Because of that, custom pointer types
> whose destructor may invalidate the pointee (i.e. they are more similar
> to Box<T> than &T) cannot internally use a reference, and must instead
> use a raw pointer.
>
> This issue is something that is often missed during unsafe review. For
> examples, see [2] and [3]. To ensure that people don't try to simplify
> RevocableGuard by changing the raw pointer to a reference, add a comment
> to that effect.
>
> Link: https://perso.crans.org/vanille/treebor/protectors.html [1]
> Link: https://users.rust-lang.org/t/unsafe-code-review-semi-owning-weak-rwlock-t-guard/95706 [2]
> Link: https://lore.kernel.org/all/aEqdur4JTFa1V20U@google.com/ [3]
> Signed-off-by: Alice Ryhl <aliceryhl@google.com>

Thanks!

Reviewed-by: Benno Lossin <lossin@kernel.org>

---
Cheers,
Benno

> ---
>  rust/kernel/revocable.rs | 4 ++++
>  1 file changed, 4 insertions(+)